MemoryCherish vs VanceAI: One Automated Pass, or Four Hours by Hand?

Take the vendor at their word first, then work out what the word means for your photograph. VanceAI states its proprietary models are powered by deep convolutional neural networks trained on millions of photos. On the restoration page specifically, VanceAI states its restorer model was trained on millions of vintage photographs to remove scratches, sepia tones, tears, cracks and spots, and that the restoration runs in one pass with no manual brushwork required.

Their published damage list is broad. VanceAI lists scratches, tears, creases, water stains, sepia and yellowing, fading, mold spots, dust and cracked emulsion as the damage types its restorer handles, and states it removes scratches, tears, stains and creases from old photos in a single automated pass. They also state the tool chains restoration, colorization and upscaling in one workflow, which is a real convenience if you want all three and you want them in one sitting.

On speed, VanceAI's site states that a heavily damaged photo restores in under ten seconds. Read that as a claim about processing, because processing is what it describes.

None of this is a knock. A single automated pass across a broad damage list, in one workflow, is a genuinely useful product for a certain kind of photograph, and there are plenty of photographs it is right for. What no product page can tell you, theirs included, is how your specific print will behave. A published capability list describes categories of damage. Your photo has a quantity of surviving information, which is a different variable, and it is the one that decides the result.

This is where comparisons usually go sideways, so let us be precise. VanceAI's site states a heavily damaged photo restores in under ten seconds. That figure describes processing: how long the software takes to produce output once the upload finishes. MemoryCherish's roughly four hours is human labor: assessment, structural repair, face work, grain matching, then your revisions. One number describes a machine running. The other describes a person deciding. They belong in different columns, and putting them in the same one produces a comparison that means nothing.

What four hours actually buys is decisions. Is that white mark dust or a highlight on a brooch? Is that dark line a crease or the strap of a bag? Does the missing corner contain part of a hand, and if it does, what position was the hand in? Those are interpretation problems rather than rendering problems, and every wrong answer gets baked permanently into the file your family will use from here on.

There is also a difference in what happens after output exists. Automated processing hands you a result. Hand work hands you a result plus a person who will change it. At MemoryCherish, revisions stay free until the picture is right, which in practice means the job ends when you say it does rather than when we say it does.

If your photo is lightly damaged and structurally complete, the fast automated path is a rational choice and we would not talk you out of it. If pieces of a face are missing, four hours of interpretation is the thing you are actually paying for.

These pricing models do not line up, so we will not force them into a table. Here is each one on its own terms.

VanceAI's pricing is credit-based, with credits giving access to AI features across its web app and its PC app. At the time of writing, their pricing page lists the AI Photo Restorer at 4 credits per image up to 4K and 8 credits at 4K. We are not converting that into a per-photo dollar figure, because what a credit is worth depends on the plan a customer holds, and any figure we printed would be our arithmetic presented as their price. At the time of writing, VanceAI states sign-up credits cover three free restorations before you choose a plan, that the free tier requires no credit card and applies no watermark, and that the web app restores one photo at a time on the free plan. They publish a 7-day money-back guarantee, state that unused subscription credits roll over each month as long as the subscription remains active, and state subscriptions can be cancelled at any time from the account dashboard. Their pricing page also states photos must be under 34 megapixels, 10MB and 8000 pixels on the longest side. On data handling, VanceAI states uploads are encrypted in transit and permanently deleted within 24 hours.

MemoryCherish charges from $38 per photo and that is the whole model. The artist tier you pick sets the figure, never the state of the print, so nobody reviews your upload and quotes upward. There are no credits to budget and no plan to remember in March. We are not the low-cost choice in this comparison and we do not present ourselves as one. What the money buys is a slice of a person's working day spent on your photograph, plus the right to send the result back until it is right.

The published damage list is not really the variable. Scratches, creases, water stains, mold spots and fading are categories, and plenty of tools handle all of them competently. The variable that decides your outcome is how much of the photograph still exists underneath the damage.

Think of it as an information problem. A scratch across a sky is easy, because the sky on either side tells you exactly what belongs in the gap. A scratch across an iris is hard, because the missing pixels were the only record of a detail unique to one person. As a category, automated restoration fills that kind of gap with statistically plausible detail rather than recovered detail, and nothing in the finished file marks which pixels came from the photograph and which were generated.

Hand work approaches the same gap differently. The artist looks for the answer before constructing one. Is the other eye intact? Is there a second photograph of the same person from a similar angle? Does the surviving edge of the tear show the direction a hairline ran? MemoryCherish artists recover what is in the photograph and never invent faces, which sometimes means telling a customer that a piece of their picture is genuinely gone. That is an uncomfortable email to send and it is the right one.

So the sorting rule is simple enough. Damage over background, texture or clothing is a good fit for automation. Damage over a face, particularly a face nobody living can double-check, is worth a human, and worth checking that you can send the result back if it comes home wrong.

Pick VanceAI if your photo is moderately damaged and structurally intact, you want restoration, colorization and upscaling handled in one workflow, you would rather self-serve than brief a person, or you want to test the water at no cost. At the time of writing they state sign-up credits cover three free restorations, with no credit card and no watermark on the free tier, so the trial costs you about ten minutes.

Pick MemoryCherish if the print has lost material, if the face is the point, if the file is destined for a frame, a memorial, a family history book or an archive donation, or if you want a named person accountable for the result. From $38, roughly four hours of hand work by US artists, revisions free until you are satisfied, cash back if we cannot get there. Our artists carry more than 37 years of combined experience, we have done work for museums and historical societies, and your original print never leaves your house.

Consider Topaz Photo if you would rather own the tools than buy an outcome. Topaz describes it as a desktop-based, AI-powered application with a set of AI-powered tools plus an autopilot mode on Mac and Windows, sold on recurring subscription plans rather than per photo, and states it offers private local rendering in which your files never leave your computer, alongside cloud rendering. That suits somebody with many photographs, some patience and an appetite for learning software.

Sequencing beats loyalty here. Run the free automated pass, look hard at the faces at full size, and spend real money only on the photographs that came back wrong.

Every VanceAI and Topaz statement above was published on those companies' own pages and checked at the time of writing in August 2026. Credit costs, free-tier terms and subscription structures move, so re-check anything you plan to budget around.

• VanceAI model description, single-pass restoration, damage list, chained workflow, under-ten-seconds claim and free-tier terms: https://vanceai.com/old-photo-restoration/
• VanceAI credit costs, money-back guarantee, credit rollover, cancellation and upload limits: https://vanceai.com/pricing/
• VanceAI neural network description and data handling statement: https://vanceai.com/
• Topaz Photo product description, archival use case, subscription model, refund window and Personal-plan commercial-use definition: https://www.topazlabs.com/topaz-photo

The omissions are deliberate. We publish VanceAI's credit costs but no dollar equivalent, and no Topaz dollar figure at all, because both are sold on subscription plans rather than per photo. Reducing either to a per-photo price would mean doing arithmetic on our side and presenting the answer as theirs. We would rather give you the model and let you read the current numbers on the page yourself.

We also describe no company's method beyond what that company publishes about its own product, because that is the part we can point to. What happens inside somebody else's software is not, and we are not going to guess about it to win an argument on our own website.